Problems says when a vector search needs an index
Problems already warned when a sort, a range or a text search over a model of 50,000 records or more ran slowly without an index. Now a nearest search on a vector field that takes a second or more without a vector index is warned about too, naming the entry to add, { fields: ["embedding"] }, so you can declare it before searches get near the time limit. It clears once the index is built, and the deploy plan lists production's.
More in Indexes.